Crop models are increasingly used in climate change impact research and integrated assessments. For the Agricultural Model Intercomparison and Improvement Project (AgMIP), 14 global gridded crop models (GGCMs) have supplied crop yield simulations (1980–2010) for maize, wheat, rice and soybean. We evaluate the performance of these models against observational data at global, national and grid cell level. We propose an open-access benchmark system against which future model versions can be tested.